JP2003536137A - Situation awareness processor - Google Patents

Abstract

(57) [Summary] A plurality of events (S) representing a situation in which the host vehicle (16) is driven are selected, including at least one set of related events (S _η ). The input data (e _i ) is a first set of data representing a target (18) in the field of view of the vehicle (16), a second set of data representing the position or movement of the vehicle (16), or Any of a third set of data representing the environment (30) of the vehicle (16) is provided to the inference engine. The inference engine operates in accordance with the inference method to respond to the input data (e _i ) and possibly one or more outputs (p _k-1 ) at a past time (k-1) to the set of inferences. An output is generated that represents the probability of occurrence of at least one of the events (S _λ ). Solution (34) may be invoked in response to one or more outputs (p _k ) from one or more inference engines.

Description

Detailed Description of the Invention

It is possible to detect and identify the environment of the own vehicle with sufficient range and accuracy, thereby avoiding a collision or injuring an occupant of the own vehicle or a pedestrian outside the own vehicle due to the collision. There is a need for an improved predictive collision sensing or collision avoidance system for automotive applications, in which appropriate measures can be selected and implemented early enough to reduce The term predictive collision sensing system as used in this application also means a collision avoidance system, which detects and tracks a target in the environment of the vehicle and then presents a measure to improve safety or Start automatically.

In general, predictive sensing systems track the movement of the vehicle relative to its environment or vice versa, for example using a radar system with an associated target tracker. The environment includes both static and moving targets. The vehicle environment differs from other target tracking environments, such as aircraft or ships, in that the vehicle is primarily driven in a road restrained environment. Of course there are exceptions to this, for example parking or off-road driving conditions,
These exceptions generally account for a relatively small percentage of vehicle operating time, or have a relatively low collision risk that would benefit from predictive collision sensing.

Referring to FIGS. 1-3, a situational awareness system 10 is shown within a radar processing system 12 that processes radar data from a radar system 14 incorporated into a vehicle 16. Generally, the vehicle 16 incorporates a sensor for detecting a target 18 in its environment and the vehicle 16 for one or more targets.
A system for activating and / or controlling the associated measures in response to the relative movement of the. The sensor for detecting the target 18 is, for example, a radar sensor system or a lidar sensor system that detects and tracks the position of the target 18 with respect to the vehicle 16 and is a co-owned U.S.A. Similar to the one disclosed in Japanese Patent No. 6085151, it is predicted whether the own vehicle 16 and the target object 18 may collide.

Referring to FIG. 1, an own vehicle 16 is moving in an X direction at an own speed 20 in an environment including one or more targets 18, and each target is associated with the own vehicle 16. At an angle 22 and a distance 24, the target velocity 26 is associated with the associated heading 28. Generally, the target object 18 is the environment 3 of the vehicle 16.
For 0, it may be stationary or moving. FIG. 2 shows a block diagram of the radar processing system 12, with each block of the block including associated software for receiving, preparing and / or processing the data provided by the radar system 14 mounted on the vehicle. . Radar system 14
The data from is pre-processed by a preprocessor 32, which produces radar data suitable for target tracking, such as range, rate of change of distance, azimuth, etc. Radar Radar generally covers a relatively wide field of view in front of the vehicle, at least ± 5 ° from the longitudinal axis of the vehicle, and depending on the radar and associated antenna configuration, may extend to ± 180 °. The system shown in this example has a field of view of ± 55 °.

The tracker 34 converts the radar output data (distance, rate of change in distance and azimuth angle) into XY coordinates that define the velocity of the target and the position of the target. For example, US Pat. No. 6,085,151 discloses a system for tracking multiple targets and forming related radar data into clusters for a single target. The associator 36 correlates old tracking data with those from recent scans and edits the tracking history of each target. The tracking map generator 38 generates a tracking map including a traveling direction and a grid of quality information from the tracking data as a record of the progress of the target object movement with respect to the own vehicle. The tracking map is updated with tracking data from subsequent targets 18 entering the field of view of the vehicle 16 and old data disappears from the map over time. Thus, the tracking map provides a representation for the vehicle 16 of the path followed by the target 18, which path is typically bound by the associated road in the environment 30 of the vehicle 16. For example, “Track Map Genera, which was separately applied as of June 8, 2001,
US patent application Ser. No. 09 / _____
The tracking map generator 38 is disclosed in _ issue.

The situation recognition processor 10 has 1) a tracking map, 2) the position of the own vehicle 16 and / or
Alternatively, the acquired data indicating the movement, that is, the own vehicle information 40 and, in some cases, 3) the environmental data 42, is used to determine the most probable or appropriate driving situation from a set of possible driving situations. to decide. For example, the environmental data 42 includes navigation data from the navigation system 44, digital maps, real-time wireless input of highway shapes and nearby vehicles, data from real-time transponders such as electromagnetic or optical markers built on highways. Alternatively, an estimate of road curvature may be provided to the situation awareness processor 10 for use in the road curvature estimator 46, along with target tracking data from the tracker 34 / associator 36. For example, navigation data such as the position and direction of the vehicle is measured by a GPS receiver, or the traveling direction of the vehicle is measured by a compass or a direction gyroscope, and the distance and the traveling direction of the vehicle are measured by measuring the speed or rotation of the wheels. Dead Reckoning Using Bearing (dead
reckoning) or a combination of the two.

The situational awareness processor 10 stores and describes the tracking map from the tracking map generator 38 and compares the time course of several tracking. By estimating the relative position and progress of the tracking target, for example, position situation, traffic situation, driving situation,
Alternatively, it becomes possible to identify various driving situations such as occurrence of a sudden event. Several techniques can be used to identify situations, such as set theory reasoning (eg, random set theory or evidence reasoning).
g)), Bayesian inference, or neural networks. Examples of location situations are divided highways or non-separated highways, intersections, highway entrances or exits, parking or off-highway situations, stationary objects to the left or right of the vehicle. As an example of traffic situation,
There are crowded traffic, quiet traffic, and normal traffic. Examples of the driving operation status include a target interruption, a lane change or speed change of the host vehicle, or a target speed change.

The situation estimated by the situation awareness processor 10 uses both the crash and crash damage severity estimates from the crash estimator 48 and crash severity estimator 50 as inputs to the response generator 52. Then, the appropriate countermeasures 54 are selected using, for example, a decision matrix. The specific response determination by the response generator 52 may be based on, for example, a rule-based system (expert system), a neural network, or another determination means.

An example of a countermeasure 54 that can be activated is an alarm device that alerts the driver to take corrective action, such as a three-dimensional audible alarm (eg, as disclosed in commonly owned US Pat. No. 5,979,586, incorporated herein by reference). Various means, such as engine throttles, vehicle transmissions, vehicle braking systems, or vehicle steering systems, for performing avoidance maneuvers to avoid collisions, as described above.
And there are various means for mitigating injury to the occupant, such as an automatic safety belt pretensioner or internal or external airbags, if a collision is unavoidable. The particular one or more countermeasures 54 selected and the method of activating, operating or controlling the one or more countermeasures 54 are described in the context awareness processor 2.
4 and the estimates of crash and crash severity.
By way of example, one possible situation is that the response to your vehicle entering the driving lane is that the target is approaching from the opposite direction or traveling in the same direction as your vehicle. , Different responses may be required depending on whether the vehicle has interrupted the road. By taking into account the traffic situation that poses a risk, the countermeasures 54 can be adapted to reduce the risk. Avoiding collisions by implementing countermeasures 54 prior to the actual collision by detecting targets within range of the vehicle using a radar system, or typically a predictive collision sensing system. Or, occupant injury due to a collision can be reduced.

Referring to FIG. 3, the radar system 14 and the radar processing system 12 are incorporated in the host vehicle 16, and the host vehicle information 40 provides, for example, the generated host vehicle information 40 to the situation recognition processor 10. Provided by an associated sensor operably connected to the vehicle processor 56. For example, one or more wheel speed sensors 58 can provide wheel speed measurements from one or more associated wheels, and the vehicle speed can be calculated from the average wheel speed for laterally opposed wheels. Is obtained, and the yaw rate is obtained from the difference in wheel speed. Alternatively, or in addition, vehicle speed may be measured from rotation of the transmission output shaft. A steering angle sensor 60 operably connected to the steering shaft provides a measure of the steering angle as the driver rotates the steering wheel. The yaw rate of the host vehicle 16 may alternatively or additionally be detected by the gyro 62, and integrating the yaw rate provides a measure of the heading of the host vehicle 16.

The situational awareness processor 10 can translate the sensor information collected by the various sensors into an understanding that is consistent with the driving environment and trigger a more appropriate response in response to an emergency. In other words, situational awareness processor 10 interprets the meaning of the sensor data. Input to the situational awareness processor 10 is translated from the "state" domain to the "event" domain. Variables in the state domain are usually represented numerically, whereas variables in the event domain are generally represented as events and by the reliability of their occurrence.
The situational awareness processor 10 identifies the "event" that corresponds to a particular "situation." Table 1
Examples of various events and associated definitions are provided in.

Inputs for identifying related events are obtained by processing original sensor information such as radar information, vehicle speed and yaw rate, GPS vehicle position information and digital map information. These inputs include, but are not limited to, Table 4
The information is interpreted by the coordinate system of FIG. 1, except for the absolute position of the vehicle, the road type, and the approach road structure. The own vehicle absolute position is defined by the coordinate system of the digital map, and the road type and approach road structure are represented by events.

[Table 1]

[Table 2]

[Table 3]

[Table 4]

The output of the situational awareness processor 10 includes a list of identified events and associated confidence levels. This event as a whole defines the current driving situation. Confidence can be expressed in several ways, such as probability and possibility, each generated by a different inference mechanism, described below. The output is
Obtained by an inference process, which is a recursive decision making process that associates an event with a corresponding input.
A decision is given by the confidence in the occurrence of a particular event. A general formula for calculating the reliability of an event is given below. Where p _k is the confidence level and e _i is the input at time k.

Various methods can be used in an inference engine that defines a recursive update mechanism. Among the most widely used are 1) Dempster-Schafer evidence reasoning, 2) fuzzy logic reasoning, 3) Bayesian reasoning, and 4) neural networks. Irregular set theory can also be used as a more general alternative to evidence or fuzzy logic reasoning. In preparation for using a particular inference engine, the above-mentioned events are first grouped into classes, each containing a certain number of mutually exclusive atomic events. The inference is performed within each class, although the output of one inference engine can also be fed back as the input of another inference engine. The decision about the nature of the current driving situation is made as a function of the collective output of all inference engines. For example, Table 5 lists a set of event classes reorganized based on the events listed in Tables 1-3.

The reorganized event-based decision process essentially forms a networked inference engine, which is shown generally in FIG. 4 and in more detail in FIG. Next, as an example, the decision process will be described by deriving the Bayesian inference of the approach road structure. The Bayes theorem is as follows.

All the information available to the inference engine , And let {e _k } be the collection of all available inputs, e _a refers to all the information that the tracking map conveys as _a whole, which in turn is a set of G grids, ie Can be disassembled into and Are the heading and the quality information of the grid ξ, respectively. For N targets, e _e = N. Superscripts can be used to distinguish targets, for example: and Indicates the speed, traveling direction, and position of the target object r (r = 1, 2, ... N). Each of the target information Defines the position of the target in X and Y coordinates, Becomes Road types and approach road structures are events in nature. Each event belongs to a class of events that may have been defined previously. In general, superscript letters are also used to distinguish one event from another in its class, eg and Specifies the road type and approach road structure, respectively.

[Table 5]

The output of the inference network contains 7 events from the 7 classes defined above, where the events are represented by S, the subscript indicates the class of the event, and the superscript is the subscript. Used to represent the desired event of the specified class. Therefore, Is an output event of the class η and an event λ of the class. For example, η = 1,
When λ = 2, Means the freeway entrance. Defined as above, the Bayesian inference for approach road structure can be derived as follows. In the Bayesian framework, the result of inference is the posterior probability of a given event (
posterior probability). The probability of trying to estimate the approach road structure is And given the collection of evidence {e _k } and relevant knowledge of class S ₂ , Next to here Is an index of events belonging to the road type class.

Current Obtained earlier to approximate By using, the following recursive probability update equation is obtained. here, Is the event and Is the probability density of the set {e _k } under the condition that is true. Considering each element of {e _k }, it becomes as follows. Here, the probability density with each condition is as follows. here, And In Case of, Becomes

[0019] A suitable method for modeling is a method that uses a Gaussian mixture distribution, that is, an approximation method that includes the sum of several Gaussian distributions as follows. here, Is the number of features to select for the approach road structure. For example, for an intersection,
The following parameters may be selected. , Μ ₁ = 0 °, μ ₂ = 90 °, μ ₃ = 90 °, σ ₁ = σ ₂ = σ ₃ = 10 ° distribution, May be modeled as follows as a uniform distribution. Can be treated like.

[0020] Event type input The probability of can be calculated as follows. here, Can be obtained by a GPS / map matching quality measure. Is Can be treated like. The actual form and parameters of the above stochastic term are simplified and empirically determined in order to reduce the computational load.

The above inference engine can be replaced by any of the other three widely used inference techniques: fuzzy logic inference, Dempster-Schafer evidence inference, and neural networks. In particular, small events (atomic ev
The collection of ents) can be set based on the class of events to be recognized, as shown in Table 6, for example.

[Table 6]

It should be noted that the set minute event is not exactly the same as the situation. Some micro-events are the intersection of several events, the events are united, part of a situation, and some are undefined. However, these small sets can collectively represent the situations defined above. Once the set of tiny events is defined, the output events can be obtained by combining the tiny events as shown in Table 7. The Dempster-Schafer mass function assigned to the event can be obtained empirically. Also, the update of the block function can be performed based on the evidence reasoning theory.

[Table 7]

Although specific embodiments have been described in detail in the foregoing detailed description and illustrated in the accompanying drawings, those of ordinary skill in the art will be directed to these details by the general teaching of the disclosure. It will be appreciated that various modifications and alternatives can be devised. Therefore, the particular arrangements disclosed are for the purpose of illustration only and are not intended to limit the scope of the invention, which is intended to be the broadest scope of all appended claims and their equivalents. Should be given.

[Brief description of drawings]

[Figure 1]   It is a figure which shows the target object seen from the own vehicle in the local coordinate system of the own vehicle.

[Fig. 2]   It is a block diagram which incorporated the situation recognition processor.

[Figure 3]   It is a block diagram which incorporated the situation recognition processor.

[Figure 4]   It is a figure which shows the whole reasoning process of a situation recognition processor.

[Figure 5]   It is a figure which shows the example of the inference process of a situation recognition system.

─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Ebring, James P.             Ann, United States Michigan 48105               Arbor, Jennings Road 6940 F-term (reference) 2C032 HB05 HB22 HC08 HD03 HD07                 2F029 AA02 AB07 AC02 AC14 AC16                 5H180 AA01 AA21 BB19 CC14 EE02                       FF04 FF22 FF32 LL01 LL04                       LL06 LL17

Claims (15)

[Claims] 1. A step of: selecting a plurality of events, each event of the plurality of events representing a situation in which a host vehicle is driven, at least two of the events being different from each other; b. B) selecting at least one set of said events related to the situation in which the host vehicle is driven; and c) a set of data, a first set of data, a second set of data, and a third set of data. Providing from at least one of the sets of data, the first set of data representing an object in the field of view of the vehicle and the second set of data being the position or movement of the vehicle. And d) selecting the inference method for the first inference engine, wherein the inference engine has at least one input. One output and And e) selecting, as a first input to the inference engine, at least one element of the first set of data, the second set of data, and the third set of data. F) generating a first output from the first inference engine in response to the first input, the first output being at least one of the set of the events.
And a step of representing a probability that one event will occur. 2. A method for identifying a situation in which a vehicle is driven according to claim 1, wherein the plurality of events relate to the position of the vehicle. 3. Each of a plurality of events related to the position of the own vehicle is driving the own vehicle in the vicinity of a separation type or non-separation type highway, driving the own vehicle in the vicinity of an intersection,
Select from driving your vehicle near the freeway, driving your vehicle near the entrance or exit to the freeway entrance, driving your vehicle near the parking lot, or driving your vehicle in an off-highway environment The method for identifying a situation in which the host vehicle of claim 2, which is a situation in which the vehicle is driven. 4. The method of identifying a situation in which a vehicle is driven according to claim 1, wherein the plurality of events relate to traffic situations. 5. Each of a plurality of events related to traffic conditions includes driving the own vehicle in an environment selected from crowded traffic, off-peak traffic, and normal traffic, and moving or stationary objects on the left or right side of the own vehicle. And a moving object or a stationary object in a lane of the host vehicle, the method of recognizing a driving situation of the host vehicle according to claim 4. 6. The method of identifying the driven situation of a host vehicle of claim 1, wherein the plurality of events relate to changes in the relative position of the target object with respect to the host vehicle. 7. An object in which each of a plurality of events related to a change in the relative position of a target object with respect to the host vehicle interrupts the lane of the host vehicle from the left side or the right side of the host vehicle, the host vehicle changing the lane to the left or right, Traffic speed change on the left or right side of the host vehicle, traffic speed change on the same lane as the host vehicle, speed change of the host vehicle, traffic redirection to the left or right, and vehicle redirection to the left or right 7. A method of identifying the driving situation of a host vehicle according to claim 6, selected from: 8. At least two events of the set of events are mutually exclusive.
A method for identifying a situation in which the host vehicle according to claim 1 is driven. 9. The first set of data is selected from the number of targets in the field of view of the vehicle, the heading of the target, the velocity of the target, the position of the target with respect to the vehicle, and the data of the first set is selected. The method of claim 1, wherein at least some of the data is generated by a tracker operably connected to a radar system. 10. The method for identifying the driving situation of a host vehicle according to claim 1, wherein the second set of data is selected from the speed of the host vehicle, the yaw rate of the host vehicle, and the position of the host vehicle. 11. The method of identifying a driven situation of a host vehicle of claim 10, further comprising the step of providing a reading of the position of the host vehicle from the navigation system. 12. A third set of data includes a tracking map representing a complex route of a target in the environment of the vehicle, a curvature or type of a road on which the vehicle is driven, a fixed structure in the field of view of the vehicle. The vehicle of claim 1, wherein the tracking map is generated from data represented by a digital map of a road surface and the tracking map is generated by a tracking map generator from data generated by a tracker operably connected to a radar system. How to identify the driving situation of a car. 13. The vehicle according to claim 1, wherein the inference method is selected from a Dempster-Schafer evidence inference method, a fuzzy logic inference method, a Bayesian inference method, a neural network, and an inference method based on irregular set theory. How to identify the driving situation. 14. A) selecting an output from an inference engine as a second input to the inference engine, said output being a past one for the first input; and b). Generating a output in response to the first and second inputs from the first inference engine. 15. The method of identifying a driven situation of a vehicle of claim 1, further comprising the step of selecting a response in response to at least one output of at least one inference engine.